AI governance

The EU classifies risk in statute, the United States governs through consumer protection and state law, China requires filing before the service goes public. Three orders, one table, with the provision visible in every row.

The question this page answers

We put an AI service into production in three markets, which classification, which registration and which human oversight applies in the EU, the United States and China?

Registers read: eu-acts · us-acts · cn-acts

Comparison

The difference table sets the requirements against each other row by row: deadline, penalty and supervision in each jurisdiction.

Requirement	EU	United States	China
Classification of the system	AI-förordningen art. 6 och bilaga III Classification follows the use: prohibited practice, high risk under Annex III, limited risk with information duties, or minimal risk. Deadline: Before the service is made available to the public Penalty: Up to EUR 15 million or three percent of global turnover Supervision: Nationell marknadskontrollmyndighet och AI-byrån Open the row in the register Read 2026-08-22	NIST AI Risk Management Framework NIST AI 100-1, funktionen MAP Federal law has no binding classification. The framework is voluntary, but is used as a yardstick in enforcement and procurement. Deadline: No deadline stated Penalty: No federal counterpart, the requirement follows from state law Supervision: National Institute of Standards and Technology Open the row in the register Read 2026-08-22	Interim Measures for Generative AI Services art. 2 och 6 The measures reach services offered to the public in China, and services with public opinion capability must undergo a security assessment. Deadline: Before the service is made available to the public Penalty: Order to rectify, warning, suspension of the service and a fine Supervision: Cyberspace Administration of China Official source Read 2026-08-22
Registration and filing	AI-förordningen art. 49 och 71 High-risk systems are registered in the EU database before being placed on the market or put into service. Deadline: Before the service is made available to the public Penalty: Up to EUR 15 million or three percent of global turnover Supervision: Europeiska kommissionen och nationell myndighet Open the row in the register Read 2026-08-22	Colorado Artificial Intelligence Act Colo. Rev. Stat. § 6-1-1703 There is no federal register. State law requires notice to the attorney general on discovery of algorithmic discrimination. Deadline: 90 days from discovery Penalty: Civil penalty per violation Supervision: Colorado Attorney General Official source Read 2026-08-22	Provisions on Algorithmic Recommendations och GenAI-åtgärderna art. 24 respektive art. 17 The algorithm is filed in the regulator's register, and generative services with public opinion capability are assessed before launch. Deadline: Ten working days from providing the service Penalty: Order to rectify, warning, suspension of the service and a fine Supervision: Cyberspace Administration of China Official source Read 2026-08-22
Transparency towards the user	AI-förordningen art. 50 The person interacting with the system must be informed, and synthetic content must be marked machine-readable. Deadline: Before the service is made available to the public Penalty: Up to EUR 15 million or three percent of global turnover Supervision: Nationell marknadskontrollmyndighet Open the row in the register Read 2026-08-22	Federal Trade Commission Act 15 U.S.C. § 45 No general labelling duty. A claim about what the model does must be true and substantiated, otherwise it is a deceptive practice. Deadline: Ongoing, for as long as the processing continues Penalty: Order and civil penalty for deceptive or unfair practice Supervision: Federal Trade Commission Open the row in the register Read 2026-08-22	Measures for the Labelling of AI-Generated Synthetic Content art. 4 och 5 Generated content must carry both a visible label and a label in the file metadata. Deadline: From 1 September 2025 Penalty: Order to rectify, warning, suspension of the service and a fine Supervision: Cyberspace Administration of China Official source Read 2026-08-22
Data and training	AI-förordningen art. 10 Training, validation and test data must be governed by a data governance practice and be relevant, representative and as error-free as possible. Deadline: Before the service is made available to the public Penalty: Up to EUR 15 million or three percent of global turnover Supervision: Nationell marknadskontrollmyndighet Open the row in the register Read 2026-08-22	California AB 2013, Generative AI Training Data Transparency Cal. Civ. Code § 3111 There is no federal rule. State law requires a published description of the training data for generative systems made available in California. Deadline: From 1 January 2026 Penalty: Civil penalty per violation Supervision: Delstatens justitiekansler Official source Read 2026-08-22	Interim Measures for Generative AI Services art. 7 Training data must come from a lawful source, respect intellectual property and contain no prohibited material. Deadline: Before the service is made available to the public Penalty: Order to rectify, warning, suspension of the service and a fine Supervision: Cyberspace Administration of China Official source Read 2026-08-22
Human oversight	AI-förordningen art. 14 High-risk systems must be designed so a natural person can monitor operation, interpret output and stop the system. Deadline: Ongoing, for as long as the processing continues Penalty: Up to EUR 15 million or three percent of global turnover Supervision: Nationell marknadskontrollmyndighet Open the row in the register Read 2026-08-22	NIST AI Risk Management Framework NIST AI 100-1, funktionerna GOVERN och MANAGE The framework describes roles, escalation and shutdown as voluntary practice, and enforcement tests the outcome against consumer protection law. Deadline: No deadline stated Penalty: Order and civil penalty for deceptive or unfair practice Supervision: National Institute of Standards and Technology och Federal Trade Commission Open the row in the register Read 2026-08-22	Interim Measures for Generative AI Services art. 9 och 14 The provider carries content producer responsibility, must stop unlawful generation and report to the authority. Deadline: Ongoing, for as long as the processing continues Penalty: Order to rectify, warning, suspension of the service and a fine Supervision: Cyberspace Administration of China Official source Read 2026-08-22
Supervision and penalty	AI-förordningen art. 99 The member state designates a market surveillance authority, and the Commission's AI Office supervises general purpose models. Deadline: No deadline stated Penalty: Up to EUR 35 million or seven percent of global turnover for prohibited practice Supervision: AI-byrån och nationell marknadskontrollmyndighet Open the row in the register Read 2026-08-22	Federal Trade Commission Act och delstatlig lag 15 U.S.C. § 45 Enforcement runs through orders and settlements, and states pursue their own cases alongside the federal regulator. Deadline: No deadline stated Penalty: Order and civil penalty for deceptive or unfair practice Supervision: Federal Trade Commission och delstaternas justitiekanslerer Open the row in the register Read 2026-08-22	Interim Measures for Generative AI Services art. 21 The authority may order rectification and suspend the service, and liability under PIPL, the Data Security Law and the Cybersecurity Law applies in parallel. Deadline: No deadline stated Penalty: Order to rectify, warning, suspension of the service and a fine Supervision: Cyberspace Administration of China Official source Read 2026-08-22

Classification of the system

AI-förordningen

art. 6 och bilaga III

Classification follows the use: prohibited practice, high risk under Annex III, limited risk with information duties, or minimal risk.

Deadline:: Before the service is made available to the public
Penalty:: Up to EUR 15 million or three percent of global turnover
Supervision:: Nationell marknadskontrollmyndighet och AI-byrån

Open the row in the register

Read 2026-08-22

United States

NIST AI Risk Management Framework

NIST AI 100-1, funktionen MAP

Federal law has no binding classification. The framework is voluntary, but is used as a yardstick in enforcement and procurement.

Deadline:: No deadline stated
Penalty:: No federal counterpart, the requirement follows from state law
Supervision:: National Institute of Standards and Technology

Open the row in the register

Read 2026-08-22

China

Interim Measures for Generative AI Services

art. 2 och 6

The measures reach services offered to the public in China, and services with public opinion capability must undergo a security assessment.

Deadline:: Before the service is made available to the public
Penalty:: Order to rectify, warning, suspension of the service and a fine
Supervision:: Cyberspace Administration of China

Official source

Read 2026-08-22

Registration and filing

AI-förordningen

art. 49 och 71

High-risk systems are registered in the EU database before being placed on the market or put into service.

Deadline:: Before the service is made available to the public
Penalty:: Up to EUR 15 million or three percent of global turnover
Supervision:: Europeiska kommissionen och nationell myndighet

Open the row in the register

Read 2026-08-22

United States

Colorado Artificial Intelligence Act

Colo. Rev. Stat. § 6-1-1703

There is no federal register. State law requires notice to the attorney general on discovery of algorithmic discrimination.

Deadline:: 90 days from discovery
Penalty:: Civil penalty per violation
Supervision:: Colorado Attorney General

Official source

Read 2026-08-22

China

Provisions on Algorithmic Recommendations och GenAI-åtgärderna

art. 24 respektive art. 17

The algorithm is filed in the regulator's register, and generative services with public opinion capability are assessed before launch.

Deadline:: Ten working days from providing the service
Penalty:: Order to rectify, warning, suspension of the service and a fine
Supervision:: Cyberspace Administration of China

Official source

Read 2026-08-22

Transparency towards the user

AI-förordningen

art. 50

The person interacting with the system must be informed, and synthetic content must be marked machine-readable.

Deadline:: Before the service is made available to the public
Penalty:: Up to EUR 15 million or three percent of global turnover
Supervision:: Nationell marknadskontrollmyndighet

Open the row in the register

Read 2026-08-22

United States

Federal Trade Commission Act

15 U.S.C. § 45

No general labelling duty. A claim about what the model does must be true and substantiated, otherwise it is a deceptive practice.

Deadline:: Ongoing, for as long as the processing continues
Penalty:: Order and civil penalty for deceptive or unfair practice
Supervision:: Federal Trade Commission

Open the row in the register

Read 2026-08-22

China

Measures for the Labelling of AI-Generated Synthetic Content

art. 4 och 5

Generated content must carry both a visible label and a label in the file metadata.

Deadline:: From 1 September 2025
Penalty:: Order to rectify, warning, suspension of the service and a fine
Supervision:: Cyberspace Administration of China

Official source

Read 2026-08-22

Data and training

AI-förordningen

art. 10

Training, validation and test data must be governed by a data governance practice and be relevant, representative and as error-free as possible.

Deadline:: Before the service is made available to the public
Penalty:: Up to EUR 15 million or three percent of global turnover
Supervision:: Nationell marknadskontrollmyndighet

Open the row in the register

Read 2026-08-22

United States

California AB 2013, Generative AI Training Data Transparency

Cal. Civ. Code § 3111

There is no federal rule. State law requires a published description of the training data for generative systems made available in California.

Deadline:: From 1 January 2026
Penalty:: Civil penalty per violation
Supervision:: Delstatens justitiekansler

Official source

Read 2026-08-22

China

Interim Measures for Generative AI Services

art. 7

Training data must come from a lawful source, respect intellectual property and contain no prohibited material.

Deadline:: Before the service is made available to the public
Penalty:: Order to rectify, warning, suspension of the service and a fine
Supervision:: Cyberspace Administration of China

Official source

Read 2026-08-22

Human oversight

AI-förordningen

art. 14

High-risk systems must be designed so a natural person can monitor operation, interpret output and stop the system.

Deadline:: Ongoing, for as long as the processing continues
Penalty:: Up to EUR 15 million or three percent of global turnover
Supervision:: Nationell marknadskontrollmyndighet

Open the row in the register

Read 2026-08-22

United States

NIST AI Risk Management Framework

NIST AI 100-1, funktionerna GOVERN och MANAGE

The framework describes roles, escalation and shutdown as voluntary practice, and enforcement tests the outcome against consumer protection law.

Deadline:: No deadline stated
Penalty:: Order and civil penalty for deceptive or unfair practice
Supervision:: National Institute of Standards and Technology och Federal Trade Commission

Open the row in the register

Read 2026-08-22

China

Interim Measures for Generative AI Services

art. 9 och 14

The provider carries content producer responsibility, must stop unlawful generation and report to the authority.

Deadline:: Ongoing, for as long as the processing continues
Penalty:: Order to rectify, warning, suspension of the service and a fine
Supervision:: Cyberspace Administration of China

Official source

Read 2026-08-22

Supervision and penalty

AI-förordningen

art. 99

The member state designates a market surveillance authority, and the Commission's AI Office supervises general purpose models.

Deadline:: No deadline stated
Penalty:: Up to EUR 35 million or seven percent of global turnover for prohibited practice
Supervision:: AI-byrån och nationell marknadskontrollmyndighet

Open the row in the register

Read 2026-08-22

United States

Federal Trade Commission Act och delstatlig lag

15 U.S.C. § 45

Enforcement runs through orders and settlements, and states pursue their own cases alongside the federal regulator.

Deadline:: No deadline stated
Penalty:: Order and civil penalty for deceptive or unfair practice
Supervision:: Federal Trade Commission och delstaternas justitiekanslerer

Open the row in the register

Read 2026-08-22

China

Interim Measures for Generative AI Services

art. 21

The authority may order rectification and suspend the service, and liability under PIPL, the Data Security Law and the Cybersecurity Law applies in parallel.

Deadline:: No deadline stated
Penalty:: Order to rectify, warning, suspension of the service and a fine
Supervision:: Cyberspace Administration of China

Official source

Read 2026-08-22

View the difference table

Where the jurisdictions diverge

The EU ties the requirements to a statutory classification, while the United States has no federal classification and instead tests claims and outcomes against consumer protection law.
China is the only regime where the service is filed before it goes public, and where labelling of generated content is an explicit duty with a technical specification.
Training data requirements exist in all three, but with different addressees: data governance in the EU, a published description in California and a lawful source in China.

Connections

Sources and review

This page restates written law with the provision and the official source. It is not legal advice in an individual matter.

Next step

Three ways to put the register to work in your own practice.

Book a walkthrough Pricing and terms API and datasets

Start with your task

Litigation
Find support in a judgment
Search guiding decisions, see what became final and follow changes in the law.
In-house, deals
Map the rules in a transaction
Move from theme to act and on to the article that carries the duty.
Compliance
Assess the risk in a process
Risk scoring per legal area, with the sources behind every score.

AI governance

The question this page answers

Comparison

Classification of the system

Registration and filing

Transparency towards the user

Data and training

Human oversight

Supervision and penalty

Where the jurisdictions diverge

Connections

Sources and review

Start with your task

Find support in a judgment

Map the rules in a transaction

Assess the risk in a process